Yang Xue, Li Tianyu, Cao Yanjia, Zheng Xiaoyun, Tang Luliang
School of Geography and Information Engineering, China University of Geosciences, Wuhan, 430074, China.
National Engineering Research Center of Geographic Information System, Wuhan, 430074, China.
Sci Rep. 2025 Jul 14;15(1):25377. doi: 10.1038/s41598-025-09779-1.
Evidence shows enhanced walking environment promotes overall physical activities and further alleviates the risk of chronic diseases and mental disorders. Current walkability research is limited by traditional GIS methods that fail to capture micro-level details and human perceptions. Additionally, existing image segmentation techniques return low accuracy when extracting complex street environment features. Therefore, we developed a hierarchical evaluation framework for urban walkability with high precision image segmentation techniques, and subjective measurements on four first-level indicators (greenness, openness, crowding, safety) and their corresponding second-level indicators. An entropy weight method was constructed to quantify the indicators based on questionnaires from 120 volunteers. Furthermore, we developed Detail-Strengthened High-Resolution Network (DS-HRNet), a deep learning model that demonstrates a 15% improvement in street scene segmentation performance compared to existing models. Using the newly developed deep learning model, we analyzed 113,900 street view images in central Wuhan City, China. Our walkability results revealed spatial heterogeneity across the city, characterized by substantial disparities between adjacent areas, particularly in commercial areas. Subsequent socioeconomic analysis demonstrated that better walkability exists in areas of higher socioeconomic status but lower proportion of non-local residents. This walkability inequality may further lead to health disparities through its influence on physical activity and social interaction.
有证据表明，改善步行环境能促进整体身体活动，并进一步降低患慢性病和精神障碍的风险。当前的步行适宜性研究受到传统地理信息系统（GIS）方法的限制，这些方法无法捕捉微观层面的细节和人类感知。此外，现有的图像分割技术在提取复杂的街道环境特征时准确率较低。因此，我们利用高精度图像分割技术以及对四个一级指标（绿化、开放性、拥挤程度、安全性）及其相应二级指标的主观测量，开发了一个城市步行适宜性的分层评估框架。基于120名志愿者的问卷调查构建了熵权法来量化这些指标。此外，我们还开发了细节增强高分辨率网络（DS-HRNet），这是一种深度学习模型，与现有模型相比，其街景分割性能提高了15%。利用新开发的深度学习模型，我们分析了中国武汉市中心的113900张街景图像。我们的步行适宜性结果揭示了整个城市的空间异质性，其特点是相邻区域之间存在显著差异，特别是在商业区。随后的社会经济分析表明，社会经济地位较高但非本地居民比例较低的地区步行适宜性更好。这种步行适宜性的不平等可能会通过影响身体活动和社会互动进一步导致健康差异。
